School Management AI Agent - ML Intent & Slot Model
This model powers the semantic intent classification, entity extraction, and schema-aware SQL reasoning core of the School Management AI Chatbot.
Model Details
- Architecture: Soft-Voting Hybrid FeatureUnion Ensemble
- Feature Extraction:
- Word N-Grams (1, 3) with sublinear TF scaling
- Character Subword N-Grams (2, 5) (char_wb) for typo resistance and morphological parsing
- Ensemble Estimators:
- Multinomial Logistic Regression (=4.0$, balanced class weights)
- SGDClassifier with Log Loss (=10^{-4}$, ElasticNet penalty)
- Multinomial Naive Bayes (=0.1$ for zero-frequency smoothing)
- Training Accuracy: 100.0% Macro F1
- Domain Scope: 27 ground-truth intents grounded in PostgreSQL schema school (29 tables, 4 views)
Files in Repository
- intent_model.joblib: Serialized Scikit-Learn pipeline
- model_metrics.json: 5-fold cross-validation report and per-class classification metrics